For the past two decades, radar images from satellites have dominated the field of geophysical monitoring for natural hazards like earthquakes, volcanoes, or landslides. These images reveal small perturbations precisely, but large changes from events like big earthquake ruptures or fast-moving glaciers remained difficult to assess from afar, until now.
Sebastien Leprince, a graduate student in electrical engineering at the California Institute of Technology, working under the supervision of geology professor and director of Caltech’s Tectonics Observatory (TO), Jean-Philippe Avouac, wrote software that correlates any two optical images taken by satellite. It has proved extremely reliable in tracking large-scale changes on Earth’s surface, like earthquake ruptures, the mechanics of “slow” landslides, or defining the fastest-moving sections of glaciers that, due to global warming, have recently increased their pace.
Leprince will describe his software and results of many of its applications on December 14 at the annual meeting of the American Geophysical Union (AGU) in San Francisco. His research will also be featured in the January 1 issue of Eos, AGU’s weekly newspaper.
When the technique called InSAR, which uses radar images to reveal details about ground displacement, was introduced, it was quickly embraced. No longer did geoscientists have to rely solely on measurements made by troupes of field geologists or by ground-based devices that might not have been optimally placed. But, says Leprince, “InSAR is physically limited: it’s good for small displacements but not for large ones. The radar resolution isn’t enough to look at deformation with a large gradient.”
Using optical images to complement the radar-based InSAR technique seemed like a natural step. When Leprince began grappling with the idea in 2003, he found several baby steps had been taken. “Satellite image correlation was not a science yet, it was more like an art,” he says. The first attempts, reported in 1991, were inconclusive but promising. Since then, several teams of scientists had worked on the problem independently. Some had even developed it well enough to monitor glacier flow.
The major obstacle Leprince faced in developing optical image correlation software was that there were several steps involved but no one knew in which order to take them. “Errors came from everywhere, but where exactly?” he noted. “And we found at least one major flaw in each step.”
Three of the four main steps involve correcting geometric distortions innate to taking pictures from space and projecting them onto a surface. The first step matches coordinates of the satellite image with coordinates on the ground. “This is not new, but the approximations being made were not okay,” says Leprince. The second step describes the satellite’s position in its orbit at the time it took the photo. This is just like in everyday life–you need to know how your camera was oriented when you show off a photo you snapped. In the next step, which Leprince says people never knew they were doing wrong, the image is correctly wrapped onto topography. Finally, the images are precisely combined-or coregistered-in order to measure surface displacements accurately.
“What is important is that we identified the steps and took each one independently and did an error analysis for each step to see how errors propagated,” says Leprince. His program, which he calls COSI-Corr and which was packaged by the TO’s software engineer Francois Ayoub for official release this year, takes all of these steps automatically in just a few hours of processing time. “You start the program, you go home, you have a nice weekend on the beach, and it’s done.”
The paper describing the software Leprince developed appeared in the June 2007 issue of the journal IEEE Transactions on Geoscience and Remote Sensing. COSI-Corr can now combine any images taken by different satellite imagers from different incidence views. For example, to analyze displacement from the 1999 Hector Mine earthquake near Twenty-Nine Palms in California, Leprince correlated a SPOT 4 image with an ASTER image. This had never been done before. It takes only a few hours to process.
Using his technique, Leprince has precisely measured offset from several notable recent earthquakes, including 2005 Kashmir, Pakistan; 2002 Denali, Alaska; 1999 Hector Mine and Chi Chi, Taiwan; and 1992 Landers, California. In the case of earthquakes, the image correlation technique can be used to map in detail all fault ruptures and to measure displacements both along and across the fault. Uncertainties, typically within centimeters for 10-15-meter-resolution images, are extremely low.
The day after Leprince released his software through the TO website, he was contacted by a geologist in Canada asking how the technique could be used to study glacier flow. Radar images cannot analyze glaciers because they move too fast and ice melting poses a problem. “The tectonic application was pretty well set up and we’d tested it thoroughly,” says Leprince. “So we extended it to glaciology.” And then to other studies as well.
What’s tricky about studying glacier flow is that not only has their pace picked up in recent years due to climate change, but glaciers have a natural yearly cycle of ice gain and loss. The two signals can be discerned with cross-correlation of optical imagery. Leprince’s method was used to study Mer de Glace glacier in the Alps, which flows at around 90 meters per year. The optical images provide a full view of the ice flow field, pinpointing exactly where the glacier is moving fastest. The same approach was taken with a landslide above the Alpine town of Barcelonnette in eastern France. Benchmarks had been planted to monitor the landslide’s flow, and Leprince’s correlation methods showed that all 38 of them missed the fastest-moving region. While the landslide is moving slow now, the town will be threatened when the landslide detaches and descends rapidly.
There are many more applications for correlating optical images to monitor Earth surface changes. Caltech geologists and their collaborators began to apply it to studying dunes, which radars cannot image, after they were contacted by labs in Egypt who need information on dune migration for urban planning.
“Radar interferometry is a huge technique, but you can only measure half of the world with it. Now we can measure the other half with this technique,” comments Leprince. “The biggest thing is what’s to come.”